Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

using namespace should be scoped to prevent name clashes, see issue #1170 #1171

Merged
merged 1 commit into from
Nov 4, 2021
Merged

using namespace should be scoped to prevent name clashes, see issue #1170 #1171

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
154 changes: 90 additions & 64 deletions perf_test/blas/blas3/KokkosBlas3_gemm_perf_test.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -416,7 +416,7 @@ void __do_gemm_serial_batched_template(options_t options,
C = Kokkos::subview(_gemm_args.C, Kokkos::ALL(), Kokkos::ALL(), j);
}

SerialGemm<TransAType, TransBType, AlgoType>::invoke(
KokkosBatched::SerialGemm<TransAType, TransBType, AlgoType>::invoke(
_gemm_args.alpha, A, B, _gemm_args.beta, C);
}
}
Expand Down Expand Up @@ -445,9 +445,9 @@ template <class scalar_type, class vta, class vtb, class vtc, class device_type,
void __do_gemm_serial_batched(options_t options, gemm_args_t gemm_args) {
char a = toupper(gemm_args.transA);
char b = toupper(gemm_args.transB);
using N = Trans::NoTranspose;
using T = Trans::Transpose;
// using C = Trans::ConjTranspose;
using N = KokkosBatched::Trans::NoTranspose;
using T = KokkosBatched::Trans::Transpose;
// using C = KokkosBatched::Trans::ConjTranspose;

STATUS;

Expand Down Expand Up @@ -479,12 +479,14 @@ template <class algo_tag, class blocking_type, class device_type,
class algo_mode = void>
void __do_gemm_parallel_batched_heuristic_template(options_t options,
gemm_args_t gemm_args) {
BatchedGemmHandle batchedGemmHandle(BaseHeuristicAlgos::SQUARE);
KokkosBatched::BatchedGemmHandle batchedGemmHandle(
KokkosBatched::BaseHeuristicAlgos::SQUARE);
char a = toupper(gemm_args.transA);
char b = toupper(gemm_args.transB);
using N = Trans::NoTranspose;
using T = Trans::Transpose;
// using C = Trans::ConjTranspose;
using N = KokkosBatched::Trans::NoTranspose;
using T = KokkosBatched::Trans::Transpose;
// using C = KokkosBatched::Trans::ConjTranspose;
using KokkosBatched::BatchLayout;

STATUS;
if (a == 'N' && b == 'N') {
Expand Down Expand Up @@ -950,9 +952,9 @@ template <class algo_tag, class blocking_type, class device_type,
void __do_gemm_parallel_batched(options_t options, gemm_args_t gemm_args) {
char a = gemm_args.transA;
char b = gemm_args.transB;
using N = Trans::NoTranspose;
using T = Trans::Transpose;
// using C = Trans::ConjTranspose;
using N = KokkosBatched::Trans::NoTranspose;
using T = KokkosBatched::Trans::Transpose;
// using C = KokkosBatched::Trans::ConjTranspose;

STATUS;

Expand Down Expand Up @@ -1515,8 +1517,12 @@ void __do_gemm_armpl(options_t options, gemm_args_t gemm_args) {
uint32_t warm_up_n = options.warm_up_n;
uint32_t n = options.n;
Kokkos::Timer timer;
char transa = std::is_same<TransAType, Trans::NoTranspose>::value ? 'N' : 'T';
char transb = std::is_same<TransBType, Trans::NoTranspose>::value ? 'N' : 'T';
char transa =
std::is_same<TransAType, KokkosBatched::Trans::NoTranspose>::value ? 'N'
: 'T';
char transb =
std::is_same<TransBType, KokkosBatched::Trans::NoTranspose>::value ? 'N'
: 'T';

if (!std::is_same<default_scalar, double>::value)
FATAL_ERROR("only double scalars are supported!");
Expand Down Expand Up @@ -2229,18 +2235,20 @@ void do_gemm_serial_blas(options_t options) {
void do_gemm_serial_batched(options_t options) {
STATUS;
__do_loop_and_invoke(
options, __do_gemm_serial_batched<default_scalar, view_type_3d,
view_type_3d, view_type_3d,
default_device, Algo::Gemm::Unblocked>);
options,
__do_gemm_serial_batched<default_scalar, view_type_3d, view_type_3d,
view_type_3d, default_device,
KokkosBatched::Algo::Gemm::Unblocked>);
return;
}

void do_gemm_serial_batched_blocked(options_t options) {
STATUS;
__do_loop_and_invoke(
options, __do_gemm_serial_batched<default_scalar, view_type_3d,
view_type_3d, view_type_3d,
default_device, Algo::Gemm::Blocked>);
options,
__do_gemm_serial_batched<default_scalar, view_type_3d, view_type_3d,
view_type_3d, default_device,
KokkosBatched::Algo::Gemm::Blocked>);
return;
}

Expand All @@ -2263,26 +2271,29 @@ void do_gemm_serial_batched_parallel(options_t options) {
if (options.blas_args.batch_size_last_dim)
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<SerialBatchDim3Tag, Algo::Gemm::Unblocked,
__do_gemm_parallel_batched<SerialBatchDim3Tag,
KokkosBatched::Algo::Gemm::Unblocked,
default_device>);
else
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<SerialTag, Algo::Gemm::Unblocked,
default_device>);
options,
__do_gemm_parallel_batched<
SerialTag, KokkosBatched::Algo::Gemm::Unblocked, default_device>);
return;
}

void do_gemm_serial_batched_blocked_parallel(options_t options) {
STATUS;
if (options.blas_args.batch_size_last_dim)
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<SerialBatchDim3Tag, Algo::Gemm::Blocked,
default_device>);
options, __do_gemm_parallel_batched<SerialBatchDim3Tag,
KokkosBatched::Algo::Gemm::Blocked,
default_device>);
else
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<SerialTag, Algo::Gemm::Blocked,
default_device>);
options,
__do_gemm_parallel_batched<
SerialTag, KokkosBatched::Algo::Gemm::Blocked, default_device>);
return;
}

Expand All @@ -2293,13 +2304,14 @@ void do_gemm_serial_simd_batched_parallel(options_t options) {
options.use_simd = true;
if (options.blas_args.batch_size_last_dim)
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<TeamSimdBatchDim4Tag, Algo::Gemm::Unblocked,
default_device, Mode::Serial>);
options, __do_gemm_parallel_batched<
TeamSimdBatchDim4Tag, KokkosBatched::Algo::Gemm::Unblocked,
default_device, KokkosBatched::Mode::Serial>);
else
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<TeamSimdTag, Algo::Gemm::Unblocked,
default_device, Mode::Serial>);
__do_loop_and_invoke(options,
__do_gemm_parallel_batched<
TeamSimdTag, KokkosBatched::Algo::Gemm::Unblocked,
default_device, KokkosBatched::Mode::Serial>);
return;
}

Expand All @@ -2310,13 +2322,14 @@ void do_gemm_serial_simd_batched_blocked_parallel(options_t options) {
options.use_simd = true;
if (options.blas_args.batch_size_last_dim)
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<TeamSimdBatchDim4Tag, Algo::Gemm::Blocked,
default_device, Mode::Serial>);
options, __do_gemm_parallel_batched<
TeamSimdBatchDim4Tag, KokkosBatched::Algo::Gemm::Blocked,
default_device, KokkosBatched::Mode::Serial>);
else
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<TeamSimdTag, Algo::Gemm::Blocked,
default_device, Mode::Serial>);
__do_loop_and_invoke(options,
__do_gemm_parallel_batched<
TeamSimdTag, KokkosBatched::Algo::Gemm::Blocked,
default_device, KokkosBatched::Mode::Serial>);
return;
}

Expand All @@ -2329,11 +2342,13 @@ void do_gemm_serial_batched_compact_mkl_parallel(options_t options) {
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<SerialSimdBatchDim3Tag,
Algo::Gemm::CompactMKL, default_device>);
KokkosBatched::Algo::Gemm::CompactMKL,
default_device>);
else
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<SerialSimdTag, Algo::Gemm::CompactMKL,
__do_gemm_parallel_batched<SerialSimdTag,
KokkosBatched::Algo::Gemm::CompactMKL,
default_device>);
return;
}
Expand Down Expand Up @@ -2367,26 +2382,29 @@ void do_gemm_team_batched_parallel(options_t options) {
if (options.blas_args.batch_size_last_dim)
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<TeamBatchDim3Tag, Algo::Gemm::Unblocked,
__do_gemm_parallel_batched<TeamBatchDim3Tag,
KokkosBatched::Algo::Gemm::Unblocked,
default_device>);
else
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<TeamTag, Algo::Gemm::Unblocked,
default_device>);
options,
__do_gemm_parallel_batched<
TeamTag, KokkosBatched::Algo::Gemm::Unblocked, default_device>);
return;
}

void do_gemm_team_batched_blocked_parallel(options_t options) {
STATUS;
if (options.blas_args.batch_size_last_dim)
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<TeamBatchDim3Tag, Algo::Gemm::Blocked,
default_device>);
options, __do_gemm_parallel_batched<TeamBatchDim3Tag,
KokkosBatched::Algo::Gemm::Blocked,
default_device>);
else
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<TeamTag, Algo::Gemm::Blocked,
default_device>);
options,
__do_gemm_parallel_batched<TeamTag, KokkosBatched::Algo::Gemm::Blocked,
default_device>);
return;
}

Expand All @@ -2396,11 +2414,13 @@ void do_gemm_team_vector_batched_parallel(options_t options) {
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<TeamVectorBatchDim3Tag,
Algo::Gemm::Unblocked, default_device>);
KokkosBatched::Algo::Gemm::Unblocked,
default_device>);
else
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<TeamVectorTag, Algo::Gemm::Unblocked,
__do_gemm_parallel_batched<TeamVectorTag,
KokkosBatched::Algo::Gemm::Unblocked,
default_device>);
return;
}
Expand All @@ -2411,12 +2431,15 @@ void do_gemm_team_simd_batched_parallel(options_t options) {
if (options.blas_args.batch_size_last_dim)
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<TeamSimdBatchDim4Tag, Algo::Gemm::Unblocked,
default_device, Mode::Team>);
__do_gemm_parallel_batched<TeamSimdBatchDim4Tag,
KokkosBatched::Algo::Gemm::Unblocked,
default_device, KokkosBatched::Mode::Team>);
else
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<TeamSimdTag, Algo::Gemm::Unblocked,
default_device, Mode::Team>);
options,
__do_gemm_parallel_batched<TeamSimdTag,
KokkosBatched::Algo::Gemm::Unblocked,
default_device, KokkosBatched::Mode::Team>);
return;
}

Expand All @@ -2426,28 +2449,31 @@ void do_gemm_team_simd_batched_blocked_parallel(options_t options) {
if (options.blas_args.batch_size_last_dim)
__do_loop_and_invoke(
options,
__do_gemm_parallel_batched<TeamSimdBatchDim4Tag, Algo::Gemm::Blocked,
default_device, Mode::Team>);
__do_gemm_parallel_batched<TeamSimdBatchDim4Tag,
KokkosBatched::Algo::Gemm::Blocked,
default_device, KokkosBatched::Mode::Team>);
else
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<TeamSimdTag, Algo::Gemm::Blocked,
default_device, Mode::Team>);
options,
__do_gemm_parallel_batched<TeamSimdTag,
KokkosBatched::Algo::Gemm::Blocked,
default_device, KokkosBatched::Mode::Team>);
return;
}

// Blocked algo not yet implemented for TeamVectorGemm.
/* void do_gemm_team_vector_batched_blocked_parallel(options_t options) {
STATUS;
__do_loop_and_invoke(
options, __do_gemm_parallel_batched<TeamVectorTag, Algo::Gemm::Blocked,
default_device>); return;
options, __do_gemm_parallel_batched<TeamVectorTag,
KokkosBatched::Algo::Gemm::Blocked, default_device>); return;
} */

void do_gemm_experiment_parallel(options_t options) {
STATUS;
using TransAType = Trans::NoTranspose;
using TransBType = Trans::NoTranspose;
using BlockingType = Algo::Gemm::Unblocked;
using TransAType = KokkosBatched::Trans::NoTranspose;
using TransBType = KokkosBatched::Trans::NoTranspose;
using BlockingType = KokkosBatched::Algo::Gemm::Unblocked;

// __do_loop_and_invoke(
// options, __do_gemm_parallel_experiment1<TransAType, TransBType,
Expand Down
24 changes: 18 additions & 6 deletions perf_test/blas/blas3/KokkosBlas3_trmm_perf_test.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -272,14 +272,15 @@ void __do_trmm_serial_batched_template(options_t options,
uint32_t warm_up_n = options.warm_up_n;
uint32_t n = options.n;
Kokkos::Timer timer;
using tag = Algo::Trmm::Unblocked;
using tag = KokkosBatched::Algo::Trmm::Unblocked;

for (uint32_t j = 0; j < warm_up_n; ++j) {
for (int i = 0; i < options.start.a.k; ++i) {
auto A = Kokkos::subview(trmm_args.A, i, Kokkos::ALL(), Kokkos::ALL());
auto B = Kokkos::subview(trmm_args.B, i, Kokkos::ALL(), Kokkos::ALL());

SerialTrmm<side, uplo, trans, diag, tag>::invoke(trmm_args.alpha, A, B);
KokkosBatched::SerialTrmm<side, uplo, trans, diag, tag>::invoke(
trmm_args.alpha, A, B);
}
// Fence after submitting each batch operation
Kokkos::fence();
Expand All @@ -291,7 +292,8 @@ void __do_trmm_serial_batched_template(options_t options,
auto A = Kokkos::subview(trmm_args.A, i, Kokkos::ALL(), Kokkos::ALL());
auto B = Kokkos::subview(trmm_args.B, i, Kokkos::ALL(), Kokkos::ALL());

SerialTrmm<side, uplo, trans, diag, tag>::invoke(trmm_args.alpha, A, B);
KokkosBatched::SerialTrmm<side, uplo, trans, diag, tag>::invoke(
trmm_args.alpha, A, B);
}
// Fence after submitting each batch operation
Kokkos::fence();
Expand All @@ -315,6 +317,11 @@ void __do_trmm_serial_batched(options_t options, trmm_args_t trmm_args) {
__trans = tolower(trmm_args.trans);
//__diag = tolower(diag[0]);

using KokkosBatched::Diag;
using KokkosBatched::Side;
using KokkosBatched::Trans;
using KokkosBatched::Uplo;

STATUS;

//// Lower non-transpose ////
Expand Down Expand Up @@ -480,8 +487,8 @@ struct parallel_batched_trmm {
auto svA = Kokkos::subview(trmm_args_.A, i, Kokkos::ALL(), Kokkos::ALL());
auto svB = Kokkos::subview(trmm_args_.B, i, Kokkos::ALL(), Kokkos::ALL());

SerialTrmm<side, uplo, trans, diag, tag>::invoke(trmm_args_.alpha, svA,
svB);
KokkosBatched::SerialTrmm<side, uplo, trans, diag, tag>::invoke(
trmm_args_.alpha, svA, svB);
}
};

Expand All @@ -491,7 +498,7 @@ void __do_trmm_parallel_batched_template(options_t options,
uint32_t warm_up_n = options.warm_up_n;
uint32_t n = options.n;
Kokkos::Timer timer;
using tag = Algo::Trmm::Unblocked;
using tag = KokkosBatched::Algo::Trmm::Unblocked;
using execution_space = typename device_type::execution_space;
using functor_type =
parallel_batched_trmm<side, uplo, trans, diag, tag, execution_space>;
Expand Down Expand Up @@ -528,6 +535,11 @@ void __do_trmm_parallel_batched(options_t options, trmm_args_t trmm_args) {
__trans = tolower(trmm_args.trans);
//__diag = tolower(diag[0]);

using KokkosBatched::Diag;
using KokkosBatched::Side;
using KokkosBatched::Trans;
using KokkosBatched::Uplo;

STATUS;

//// Lower non-transpose ////
Expand Down
Loading